UNITED STATES v. DELEON (2024)

Our prior-precedent rule requires us to follow Eleventh Circuit precedent—even if we disagree with it or think that prior panels have overlooked important arguments—unless and until the Supreme Court or our court sitting en banc abrogates the precedent. This is one of those cases where the prior-precent rule controls the outcome.

The United States Sentencing Guidelines impose a two-level enhancement, as relevant here, for an armed robbery, “if any person was physically restrained to facilitate commission of the offense or to facilitate escape.” U.S. Sent'g Guidelines Manual § 2B3.1(b)(4)(B). This case requires us to determine whether the victim was “physically restrained” under the meaning of the Guidelines when Defendant Joseph Deleon walked into a store, pointed a gun at the cashier while demanding money from the register, received the money, and then left, all within about one minute.

After careful review of the record and case law, and with the benefit of oral argument, we are bound to affirm.

I. BACKGROUND

Deleon walked into a convenience store, went up to the counter, and asked for cigarettes. He handed the cashier $10 to pay for the cigarettes. But when the cashier opened the register, Deleon took out a handgun and pointed it at the cashier. A video of the offense shows that the counter put little—but some—distance between Deleon and the cashier. The cashier gave Deleon all the money from the register.

Deleon wasn't satisfied. So he repeatedly signaled for the cashier to look again in the register, including by reaching over the counter to point into the register. So the cashier kept looking. Out of cash, the cashier gave Deleon $40's worth of United States postal stamps. At no point does the record indicate that Deleon actually touched the cashier. Deleon then left the store. Based on the video, the entire incident lasted about one minute.

A grand jury indicted Deleon for Hobbs Act robbery, in violation of 18 U.S.C. § 1951(a) and (b) (Count 1), and brandishing a firearm during and in relation to a crime of violence, in violation of 18 U.S.C. § 924(c)(1)(A)(ii) (Count 2). After a trial, a jury convicted Deleon on both counts. The district court sentenced Deleon to 240 months’ imprisonment (156 months on Count 1 and 84 months on Count 2, running consecutively). Over Deleon's objection, this sentence included a two-level physical-restraint enhancement under section 2B3.1(b)(4)(B) of the Guidelines. To support this enhancement, the district court found that Deleon had physically restrained the cashier by pointing a gun at him and forcing him to empty the register.

Deleon appealed that sentence but did not challenge the physical-restraint enhancement. The Eleventh Circuit vacated the sentence and remanded for resentencing on the ground that the district court had added a career-offender enhancement even though Hobbs Act robbery is not a crime of violence for purposes of that enhancement. United States v. Deleon, 812 F. App'x 948, 952 (11th Cir. 2020) (per curiam).

On remand, the probation officer again recommended the two-level enhancement for physical restraint under section 2B3.1(b)(4)(B). Deleon again objected to the enhancement, arguing that his conduct did not rise to the level of “physical restraint.” And the district court again rejected Deleon's arguments and imposed the physical-restraint enhancement.

In support of its determination, the court explained,

[Deleon] did pull out a gun. He held it to the victim. He forced the victim to comply by passing him the goods he sought to secure, and the victim clearly in this case did not feel at liberty to leave in light of the presence of the weapon. In fact, as I recall, the victim was so entranced by and terrorized by the weapon that that was all he could remember. And so the [c]ourt finds under the circumstances of this case, the use of the weapon did ensure compliance by the victim and did impede the victim from fleeing to another part of the store and restrained the victim to the space where he was standing in order to meet the demands of Mr. Deleon.

Sent'g Hr'g 10, Sept. 15, 2021.

Then the court resentenced Deleon to 135 months in prison (51 months as to Count 1 and 84 months as to Count 2, running consecutively). Relevant here, the total offense level for Count 1 (robbery) under the Guidelines was twenty-two. That total offense level included a base level of twenty plus the two-level physical-restraint enhancement. Count 2 (using a firearm during a crime of violence) added an additional seven years under 18 U.S.C. § 924(c)(1)(A)(ii).

Deleon timely appealed.

II. DISCUSSION

On appeal, Deleon challenges his sentence based solely on the district court's application of the two-level sentencing enhancement for physical restraint under section 2B3.1(b)(4)(B). He argues that if the enhancement applies here, it necessarily applies in all armed robbery cases, effectively raising the base level of armed robbery to twenty-two instead of twenty.

We first addressed the scope of the physical-restraint enhancement in United States v. Jones, 32 F.3d 1512 (11th Cir. 1994). There, we considered the commentary. See id. at 1518. We noted that the commentary defined “physically restrained” as “the forcible restraint of the victim such as by being tied, bound, or locked up.” Id. (citation omitted). Then we reasoned that “such as” in that definition “indicates that the illustrations of physical restraint are listed by way of example rather than limitation.” Id. (citation and quotation marks omitted). With that in mind, we adopted the Seventh and Eighth Circuits’ understanding of the definition of “physically restrained”: “a defendant physically restrains his victims if he creates circumstances allowing the persons no alternative but compliance.” Id. at 1519 (citation and quotation marks omitted).

Several years later, in United States v. Victor, 719 F.3d 1288 (11th Cir. 2013), we built on what we said in Jones. There, the defendant pretended he had a gun in his pocket and pointed it at a bank employee. Id. at 1289. Then he herded the employee to the teller line and yelled that he had a gun and would kill any employee who did not comply with his demands. Id. We affirmed application of the enhancement. In doing so, we explained that “by threatening the [bank] employee with what the employee believed to be a gun to prevent her from escaping,” the defendant physically restrained her under Jones. Id. at 1290. Even though he did not actually have a gun, we reasoned, the victim believed he did “so that she was forced to comply.” Id. And when the defendant argued that he had not moved the teller far, we noted that section 2B3.1(b)(4)(b) “contains no requirement that the victim be moved at all.” Id. In short, we said, the enhancement applies when “the defendant's conduct ensured the victims’ compliance and effectively prevented them from leaving a location.” Id. (citation and quotation marks omitted).

Most recently, in United States v. Ware, 69 F.4th 830 (11th Cir. 2023), cert. denied, No. 23-5946, ––– U.S. ––––, 144 S.Ct. 1395, 218 L.Ed.2d 681 (2024), we invoked Jones and Victor and again said that section 2B3.1(b)(4)(B) applies “where a defendant creates circumstances allowing [his victims] no alternative but compliance.” Id. at 854 (citation and quotation marks omitted).

In dicta, we offered that “[w]e would have a closer case if [the defendant] had entered these establishments, pointed a gun at the victim behind the welcome counter, demanded, ‘Your money or your life,’ obtained money from the victim, and left without further incident.” Id. at 855. But under the test we again approved in Ware—the enhancement applies “where a defendant ‘creates circumstances allowing [his victims] no alternative but compliance’ ”—we don't see how the enhancement wouldn't apply to the then-hypothetical scenario we raised in Ware. Id. at 854. In that case, the robber would have created circumstances allowing his victim no realistic alternative but compliance. After all, who would think compliance was optional in that situation?

And that hypothetical situation from Ware is nearly identical to the fact pattern here. As we've noted, here, Deleon pointed a gun at the cashier and demanded the cashier provide him with money. Though he didn't say, “Your money or your life,” that message surely came through loud and clear, all the same. So Deleon “create[d] circumstances allowing [the cashier] no alternative but compliance.” And under our precedent, the district court properly applied section 2B3.1(b)(4)(B)’s enhancement for “physical restraint.”

Deleon argues that, despite our precedent, the enhancement should not apply here. First, he asserts that applying the enhancement to these facts means it will apply to virtually every armed robbery, effectively raising the base offense level to twenty-two instead of twenty. Deleon also argues that applying the physical-restraint enhancement when a defendant demands money at gunpoint during an armed robbery punishes him a second time for the same conduct as his sentence for his conviction under § 924(c)(1)(A)(ii) for brandishing a firearm during a crime of violence. In other words, Deleon makes a double-counting argument.

The short answer to both of Deleon's arguments is this: as we've explained, our precedent binds us to conclude that the enhancement applies to conduct like Deleon's, and “we have categorically rejected an overlooked reason or argument exception to the prior-panel-precedent rule.” In re Lambrix, 776 F.3d 789, 794 (11th Cir. 2015) (per curiam). End of story.

To the extent that Deleon argues that applying the physical-restraint enhancement on the facts of his particular case wrongly double-counts the harms from his crime, we must also reject that argument.

In support of his double-counting argument, Deleon points to the Guidelines commentary under section 2K2.4 that precludes, in part, application of any specific offense characteristic for brandishing or using a firearm if the defendant is also sentenced under § 924(c). He points out that the probation officer did not recommend applying the six-level enhancement for when “a firearm was otherwise used” under section 2B3.1(b)(2)(B) because of the section 2K2.4 commentary. By the same logic underlying the section 2K2.4 commentary, Deleon reasons that we shouldn't apply the physical-restraint enhancement because a § 924(c) conviction already addresses the conduct of brandishing a gun.

Double counting can be permissible or impermissible. We've explained that “[i]mpermissible double counting occurs only when one part of the Guidelines is applied to increase a defendant's punishment on account of a kind of harm that has already been fully accounted for by application of another part of the Guidelines.” United States v. Dudley, 463 F.3d 1221, 1226–27 (11th Cir. 2006) (emphasis added) (citation omitted).

In contrast, “double counting is permissible” when “(1) the Sentencing Commission intended the result; and (2) each guideline section in question concerns a conceptually separate consideration related to sentencing.” United States v. Grushko, 50 F.4th 1, 16 (11th Cir. 2022). And we “presume that the Sentencing Commission intended to apply separate guideline sections cumulatively unless we are specifically directed otherwise.” Id. For these reasons, we've recognized that the same conduct can implicate “multiple guidelines sections.” Id.; see also United States v. Suarez, 893 F.3d 1330, 1337 (11th Cir. 2018) (“It is irrelevant that [two sentencing] adjustments will often be triggered by the same conduct.”).

Looking to the guidelines at issue here, the sentence for brandishing a gun and the enhancement for physically restraining a victim do not necessarily and fully cover the same conduct. For instance, an armed robber who enters a store with a gun on his waistband and heads straight for the cash register without blocking any exit or demanding assistance with the cash register may have brandished a gun, but he likely has not “physically restrained” his victims—even under our Circuit's broad definition of the phrase.

And though the commentary to section 2K2.4 directs courts not to apply enhancements “for possession, brandishing, use, or discharge” of a firearm when a separate sentence for a § 924(c) violation applies, the Guidelines include no similar commentary directing us not to apply the physical-restraint enhancement in this circumstance. Nor do we see any other indication that the Sentencing Commission specifically directed us not to apply these distinct sentence components cumulatively. See United States v. Stevens, 580 F.3d 718, 722 (8th Cir. 2009) (“The specific offense characteristic punished by § 2B3.1(b)(4)(B) is the use of physical restraint, not the presence of a firearm. The firearm was merely a tool used to effect the physical restraint accomplished by [defendants] ․ Consequently, ․ the challenged sentencing enhancement does not amount to double counting.”); United States v. Wilson, 198 F.3d 467, 472 n.1 (4th Cir. 1999) (explaining that “ ‘double counting’ is permissible under the sentencing guidelines except where it is expressly prohibited,” and the Guidelines do not expressly prohibit a physical-restraint enhancement on top of a sentence for a § 924(c) conviction).

For all these reasons, we must reject Deleon's claim that section 2B3.1(b)(4)(B)’s physical-restraint enhancement does not apply to his conduct. And we therefore affirm the district court.

III. CONCLUSION

We AFFIRM the district court's application of the physical-restraint enhancement.

The issue in this case is ripe for en banc review. As the Majority Opinion explains, the Jones/Victor line of cases binds us to affirm the application of the physical-restraint enhancement. That's because our precedent has understood the enhancement to apply any time “the defendant's conduct ensured the victims’ compliance and effectively prevented them from leaving a location.” United States v. Victor, 719 F.3d 1288, 1290 (11th Cir. 2013) (citation and quotation marks omitted). But if I were writing on a clean slate, I would conclude that section 2B3.1(b)(4)(B) does not reach Deleon's conduct. Quite simply, a plain reading of the text of section 2B3.1(b)(4)(B) does not support our precedent's standard or application of that guideline here.

When we construe a guideline, we start with its text. See United States v. Dupree, 57 F.4th 1269, 1277 (11th Cir. 2023) (en banc). The version of section 2B3.1(b)(4)(B) that applied to Deleon's conduct (2018) provides, “[I]f any person was physically restrained to facilitate commission of the offense or to facilitate escape, increase by 2 levels.”

To determine the meaning of this phrase, we “apply[ ] our traditional tools of statutory interpretation,” id. at 1277. Only if the text remains ambiguous after that do we then look to the Guidelines commentary. Id. at 1276–77, 1279.¹

Following this roadmap, I begin with the ordinary meaning of “physically restrained,” § 2B3.1(b)(4)(B), the phrase that drives the meaning of the guideline here. To determine the ordinary meaning of a term, we often turn to dictionary definitions. In re Walter Energy, Inc., 911 F.3d 1121, 1143 (11th Cir. 2018). “Restrain” means “[t]o restrict, limit, confine.”² Other definitions include the following: “to prevent from doing, exhibiting, or expressing something,” “to limit, restrict, or keep under control,” “to moderate or limit the force, effect, development, or full exercise of,” and “to deprive of liberty.”³

All these definitions contemplate some restriction, limitation, or confinement. But none specify the type of restriction, limitation, or confinement.

That's where “physically” comes in. Section 2B3.1(b)(4)(B) qualifies the word “restrained” with “physically.” “Physically” means “[w]ith regard to the body; in bodily terms or by bodily means,”⁴ “in a physical manner: in accord with physical laws,”⁵ or “in respect to the body.”⁶ Similarly, the relevant definition of “physical,” the adjective form of “physically,” is “[o]f or relating to natural phenomena perceived through the senses (as opposed to the mind); of or relating to matter or the material word; natural; tangible; concrete.”⁷

Putting it all together, then, under the guideline, a restraint—or limitation or confinement—must concern the body for the enhancement to apply. A restraint that is solely mental or psychological in nature should not qualify under section 2B3.1(b)(4)(B)’s plain text.

Not only does a non-physical restraint fail to comport with the definition of “physically,” but construing “physically restrained” to include non-physical restraints—no matter how intimidating a non-physical restraint may be—makes “physically” meaningless in the guideline.

Indeed, if the framers of the guideline wanted it to apply whenever “any person was ․ restrained” in either a physical or non-physical way, they wouldn't have included the qualifier “physically.” But the guideline contains the modifier “physically” before “restrained.” That adverb has meaning. And by its plain meaning, “physically restrained” should not include psychologically or emotionally “restrained.” After all, “[i]t is ‘a cardinal principle of statutory construction’ ” that, “ ‘if it can be prevented, no clause, sentence, or word shall be superfluous, void, or insignificant.’ ” TRW Inc. v. Andrews, 534 U.S. 19, 31, 122 S.Ct. 441, 151 L.Ed.2d 339 (2001) (quoting Duncan v. Walker, 533 U.S. 167, 174, 121 S.Ct. 2120, 150 L.Ed.2d 251 (2001)). See Antonin Scalia & Bryan Garner, Reading Law: The Interpretation of Legal Texts 176 (2012) (“Because legal drafters should not include words that have no effect, courts avoid a reading that renders some words altogether redundant.”).

When we give every word its meaning in section 2B3.1(b)(4)(B), I understand this enhancement to reflect Congress's view that physical bounding adds another dimension to the intimidation a victim of an armed robbery endures. In other words, it's bad enough to be robbed at gunpoint, but being tied up or otherwise physically bound is invasive, dangerous, and victimizing in yet another way.

Given that the text is clear, if I were writing on a clean slate, I would end with the text as well as starting with it. See Dupree, 57 F.4th at 1276–77. And I would not apply the enhancement in Deleon's case. But even if the text were ambiguous (it's not), the application notes and commentary are wholly consistent with and fully support the plain meaning of the text. See id.  ⁸

The Application Notes define “physically restrained” as “the forcible restraint of the victim such as by being tied, bound, or locked up.” § 2B3.1 cmt. n.1 (citing § 1B1.1 cmt. n.1(L)). And separately, the commentary states that the “guideline provides an enhancement for robberies where a victim ․ was physically restrained by being tied, bound, or locked up.” § 2B3.1 cmt. background. So just as the text itself conveys, the commentary and application notes reflect that section 2B3.1 contemplates enhancement upon only the use of physical restraints.⁹

Were we to apply this understanding of section 2B3.1(b)(4)(B) to Deleon's case, I would agree with Deleon that the enhancement does not apply. Deleon pointed a gun at the cashier to obtain his compliance. I can certainly understand why the cashier did not try to leave under those circumstances. But I cannot say that the pointed gun “physically” restrained the cashier to facilitate Deleon's crime because it did not physically bind, confine, or apply any force to the cashier's body. So by its plain meaning, section 2B3.1(b)(4)(B) should not apply.¹⁰

At least five of our sister circuits construe section 2B3.1(b)(4)(B) more narrowly than we do and require some type of “physical” restraint. Take United States v. Anglin, 169 F.3d 154 (2d Cir. 1999). There, the Second Circuit held that “displaying a gun and telling people to get down and not move, without more, is insufficient to trigger the ‘physical restraint’ enhancement.” Id. at 164.

The court gave three reasons. First, it explained that the conduct there was “materially different from the Guidelines examples” of physical restraint, namely, being “tied,” “bound,” or “locked up.” Id. Second, it said that the plain meaning of “physical restraint” requires more than pointing a gun and giving directions not to move. Id. The court reasoned that the Guidelines would not have included the word “physical” in section 2B3.1(b)(4)(B) if the enhancement applies even when the victim believes they are restrained but are not actually “subjected to physical restraint.” Id. at 164–65. Third, the court found that if the enhancement applied in that case, “virtually every robbery would be subject to the 2-level enhancement for physical restraint unless it took place in unoccupied premises.” Id. at 165. In the Second Circuit's view, that would be a “problematic effect for a provision drafted to deal with a special circumstance.” Id.

Similarly, in United States v. Garcia, 857 F.3d 708 (5th Cir. 2017), the Fifth Circuit held that the physical-restraint enhancement did not apply when the defendants held firearms, one defendant pointed a gun at an employee and told them to get on the ground, and another stood by the exit. Id. at 712. The court explained that the defendants “never forced [the victims] to move to a confined space” or subjected the victims to “the type of physical restraint that victims experience when they are tied, bound, or locked up.” Id. at 713. Instead, the court reasoned that the “defendants did not do anything with their firearms that goes beyond what would normally occur during an armed robbery,” so the enhancement did not apply. Id. at 713–14. The Fifth Circuit cited Anglin with approval. Id. at 713.

The Seventh, Ninth, and D.C. Circuits also require something more than pointing a gun and giving a command to justify the physical-restraint enhancement. See United States v. Herman, 930 F.3d 872, 875–77 (7th Cir. 2019) (holding that pointing a gun at a victim and ordering them not to move does not qualify as physical restraint); United States v. Parker, 241 F.3d 1114, 1118–19 (9th Cir. 2001) (holding that the enhancement requires “more than briefly pointing a gun at a victim and commanding her once to get down to constitute physical restraint, given that nearly all armed bank robberies will presumably involve such acts”); United States v. Drew, 200 F.3d 871, 880 (D.C. Cir. 2000) (holding the enhancement did not apply even where the defendant forced the victims to walk at gunpoint because, even though the victims felt restrained, they were not physically restrained, and “[a]ny other interpretation would effectively add the two-level adjustment to almost any attempted murder because presumably any victim would feel restrained if directed to move at gunpoint”).

In contrast, by my count, four other circuits (besides ours) construe the physical-restraint enhancement broadly to apply when a defendant restricts a victim's movement by using a gun, even in the absence of bodily contact, confinement, or other physical force.¹¹ See United States v. Wallace, 461 F.3d 15, 34–35 (1st Cir. 2006) (affirming application of the physical-restraint enhancement where the defendants pointed their guns at the victims, told them not to move, and blocked one victim's path); United States v. Dimache, 665 F.3d 603, 608–09 (4th Cir. 2011) (affirming application of the physical-restraint enhancement where the defendant used a gun to force bank tellers to the floor because they “were prevented from both leaving the bank and thwarting the bank robbery”); United States v. Howell, 17 F.4th 673, 692 (6th Cir. 2021) (affirming application of the physical-restraint enhancement where the defendant ordered the victim at gunpoint to lie on the floor and threatened to kill her unless her coworkers complied with orders); United States v. Miera, 539 F.3d 1232, 1235–36 (10th Cir. 2008) (affirming application of the physical-restraint enhancement where the defendant waved a gun around, told people not to move, and blocked the bank doors).

As the circuit split deepens on the proper interpretation of the physical-restraint enhancement, Deleon's case provides a good opportunity to reexamine our jurisprudence. Given the heavier emphasis on textualism in the last several years—and especially against the background of Dupree—we should align our jurisprudence with what section 2B3.1(b)(4)(B) actually says.

Because our jurisprudence on § 2B3.1(b)(4)(B) is not faithful to the text of the physical-restraint enhancement, I respectfully submit that we should revisit it en banc.

This is a sequel of sorts to my separate opinion in Snell v. United Specialty Insurance Co., 102 F.4th 1208 (11th Cir. 2024) (Newsom, J., concurring). There, I floated the following suggestion, which I suspected (rightly, as it turns out) would be fairly provocative: “Those, like me, who believe that ‘ordinary meaning’ is the foundational rule for the evaluation of legal texts should consider—consider—whether and how AI-powered large language models like OpenAI's ChatGPT, Google's Gemini, and Anthropic's Claude might—might—inform the interpretive analysis.” Id. at 1221. With the benefit of a little perspective, and incorporating by reference here all the caveats that I expressed there, I stand by what I said.

The LLM research that I conducted in connection with this case, though, raised a question that I hadn't squarely confronted in Snell, one that seemed worth flagging and exploring: What should we make of the fact that the models sometimes provide subtly different answers to the exact same question? I'll confess that the variation initially spooked me, but it now (for reasons I'll get into) seems not only unremarkable but perhaps even expected. And significantly from an ordinary-meaning perspective, it accurately reflects real people's everyday speech patterns.

Let me explain.

I

First, a bit of background. As the majority opinion explains, the district court here enhanced Deleon's sentence on the ground that he “physically restrained” one of his victims during the course of a robbery. U.S.S.G. § 2B3.1(b)(4)(B). I agree with the majority that our decisions in United States v. Jones, 32 F.3d 1512 (11th Cir. 1994), and United States v. Victor, 719 F.3d 1288 (11th Cir. 2013), control this case and require us to affirm the enhancement. See Maj. Op. at –––– – ––––. Having said that, I also agree with Judge Rosenbaum that Jones and Victor were wrongly decided and that we should rehear this case (or another one like it) as an en banc court to overrule them. See Rosenbaum Conc. Op. passim.

In Jones and Victor, panels of this Court held that § 2B3.1(b)(4)(B)’s enhancement applies where, even in the absence of any physical contact, the defendant's possession of a gun “ensured the victims’ compliance and effectively prevented them from leaving,” Jones 32 F.3d at 1519, or the defendant “threaten[ed a victim] with what [she] believed to be a gun to prevent her from escaping,” Victor, 719 F.3d at 1290. I disagree. For reasons that I'll explain—and that complement those Judge Rosenbaum has set out in her separate opinion—neither circumstance constitutes “physical[ ] restrain[t]” within the ordinary meaning of that phrase. See Rosenbaum Conc. Op. at –––– – ––––.

One thing that makes the interpretive issue in this case interesting—and perhaps a little bit harder than usual—is that it turns on the meaning not of a single word, but rather of a composite, multi-word phrase: “physically restrained.” That poses a challenge because so far as I'm aware, the entire phrase—“physically restrained”—isn't defined in any reputable dictionary, the tool to which plain-language interpreters typically turn. Given that limitation, I think Judge Rosenbaum's concurrence nicely exemplifies the standard approach. Specifically, we break the phrase into its constituent parts, use conventional interpretive tools to gauge the ordinary meaning of each, and then piece those definitions back together into a coherent whole. See, e.g., Garland v. Cargill, 602 U.S. 406, 415–16, 144 S.Ct. 1613, 219 L.Ed.2d 151 (2024) (using the breaking-and-repiecing method to define the phrase “function of the trigger”); Barton v. U.S. Att'y Gen., 904 F.3d 1294, 1298–99 (11th Cir. 2018) (Newsom, J.) (same, defining the phrase “renders admissible”). So, Judge Rosenbaum explains, “physically restrained” means, in essence, the state of being “restrained”—i.e., “restrict[ed], limit[ed], confine[d]”—by “physical[ ]” means—i.e.,“[w]ith regard to the body.” See Rosenbaum Conc. Op. at –––– – –––– (citing dictionary definitions of both terms).¹

That seems about right to me. But because there's no ready dictionary definition of the composite phrase, because by their very nature LLMs aim to capture and reflect how real people ordinarily use both individual words and multi-word phrases, and because, well, I couldn't help myself, I queried ChatGPT.²

Here's what happened.

II

Me: “What is the ordinary meaning of ‘physically restrained’?”

ChatGPT:

The ordinary meaning of “physically restrained” refers to the act of limiting or preventing someone's movement by using physical force or some kind of device. This can involve holding someone down, using handcuffs, straps, or other mechanisms to restrict their ability to move freely. The term implies an active and physical intervention to control or confine a person's movements.

So far, so good. ChatGPT's response basically squared with what I had assumed the phrase “physically restrained” means. And as it turned out, its answer also corresponded with the result of the traditional breaking-and-repiecing method exemplified in Judge Rosenbaum's concurrence. See Rosenbaum Conc. Op. at –––– – –––– .

For reasons I've explained before, though, I didn't want to seize on ChatGPT's explanation just because it matched my priors. See Snell, 102 F.4th at 1225 (Newsom, J., concurring). So, I asked Claude ³ the same question: “What is the ordinary meaning of ‘physically restrained’?” Reassuringly, Claude's response largely mirrored ChatGPT's:

The ordinary meaning of “physically restrained” generally refers to:

1. The act of using physical force or methods to limit or control someone's movements or actions.

2. A state in which a person is prevented from moving freely or acting as they wish due to physical barriers or interventions.

This can involve various forms of restraint, such as:

• Holding someone back with one's hands or body

• Using devices like handcuffs, straps, or restraining jackets

• Confining someone to a specific space, like a locked room

The term is often used in contexts related to law enforcement, medical settings, or situations where someone's freedom of movement needs to be restricted for safety or security reasons. However, it can also be used in more general situations where physical limitation is applied.

Then, for reasons I don't specifically recall—but that can presumably be chalked up to a “better safe than sorry” instinct—I asked Claude the exact same question again. To my surprise, its answer the second time around was ever-so-slightly different:

The ordinary meaning of “physically restrained” typically refers to:

Using physical means to limit or prevent someone's movement or actions.

This usually involves:

• Holding someone back

• Using restraining devices (e.g. handcuffs)

• Confining someone to a limited space

It implies direct physical intervention to restrict a person's freedom of movement.

Hmm. Claude's responses were basically the same—but not exactly, not verbatim. The substance seemed pretty much identical, but the style and structure were just a smidgen off. Most obviously, the first was longer and more detailed than the second. For instance, unlike the first response, the second didn't include a “state in which” explanation. In the “holding someone back” bullet, the first specified a means—“with one's hands or body”—while the second didn't. Similarly, in the “devices” bullet, the first referred to handcuffs, straps, and restraining jackets as examples, while the second referred only to handcuffs. Finally, the first response concluded with a short paragraph about the various “contexts” in which the phrase “physically restrained” is often used—law enforcement, medicine, etc.—while the second closed with a more generic statement that the phrase “implies direct physical intervention to restrict a person's freedom of movement.”

Honestly, Claude's varying responses rocked me back on my heels a little bit. Whatever's going on inside the LLM's black box, I worried about the same model's failure (inability? refusal?) to give the same answer to identical back-to-back queries. Might it suggest that the LLM wasn't accurately communicating its prediction about how ordinary people use a phrase like “physically restrained”—or worse, that it didn't even really have an accurate prediction? After all, standard printed dictionaries don't evolve or change their minds—certainly not within the span of a few seconds. At least within a particular bound edition, their definitions are static.

Having done some sleuthing, I'm now less concerned than I once was. There is, I think, both a valid technical explanation for the variation and, perhaps more importantly for our purposes, a pretty powerful analogy to the way that ordinary people ordinarily use phrases in ordinary conversation. I'll unpack both the explanation and the analogy shortly. First, though, let me describe the results of a little (and I do mean little) experiment that I ran, which will show, descriptively, the sort of variability that I encountered.

III

Surprised by Claude's second, slightly different response to my query, I decided to go both a little broader and a little deeper. I queried the three leading models—GPT, Claude, and Gemini ⁴ —ten times apiece. Each time, I opened a new chat and asked the same question: “What is the ordinary meaning of ‘physically restrained’?” Not perfectly scientific, I know, but better than a rifle shot.

My hope was that my humble little mini-experiment might begin to reveal how “confident” a model is (and we can be) about its response. Cf. Yonathan Arbel & David A. Hoffman, Generative Interpretation, 99 N.Y.U. L. Rev. 451, 487–488 (2024); Christoph Engel & Richard H. McAdams, Asking GPT for the Ordinary Meaning of Statutory Terms 15–16 (Max Planck Inst. Discussion Paper No. 2024/5), https://papers.ssrn.com/sol3/papers.cfm?abstract_id=4718347 [https://perma.cc/7A2K-VLBC]. A series of wildly different answers would probably suggest (and would certainly inspire) a lack of confidence.⁵ If the model instead returned essentially consistent answers—responses that coalesced around a common core—then one could probably say, with a higher degree of confidence, that the model was getting at the “correct” response.⁶

Again reassuringly, the 30 results I received—10 apiece from each of the three leading LLMs—largely echoed the initial response that I got from ChatGPT. If you're interested in the nitty gritty, all the responses are available in the Appendix. But here's the gist: When defining “physically restrained,” the models all tended to emphasize “physical force,” “physical means,” or “physical barriers.” ChatGPT and Claude specifically used one (or more) of those phrases in every one of their responses. For whatever reason, Gemini was a little different. It didn't invariably employ one of those terms explicitly, but even when it didn't, the concept of what I'll call corporeality (via either human touch or a tangible object) pervaded and tied together its example-laden answers.⁷

To be sure, the models’ responses exhibited some minor variations in structure and phrasing. ChatGPT's answers, for example, tended to fluctuate in length by a sentence or two. For its part, Claude altered the number of examples it provided from one response to the next. But for reasons I'll explain in the next part, these subtle, marginal divergences were probably (and should have been) expected. Far more importantly, I think, the responses did coalesce, substantively, around a common core—there was an objectively verifiable throughline. For our purposes, what matters is that the LLMs consistently defined the phrase “physically restrained” to require the application of tangible force, either through direct bodily contact or some other device or instrument. And that, again, squares comfortably with the results obtained through the traditional, dictionary-driven breaking-and-repiecing method.

IV

So, what to make of the slight variations among the answers that the models returned in response to my query? For present purposes, I think there are two important points. First, there's a technical explanation for the variation, which, upon reflection, doesn't much concern me—or, upon further reflection, even much surprise me. Second, there is, upon even further reflection, a sense in which the substantively-identical-and-yet-marginally-different answers (perhaps ironically) underscore the models’ utility in the ordinary-meaning analysis—namely, in that they pretty closely mimic what we would expect to see, and in fact do see, in everyday speech patterns.

A

First, the technical details—so far as I understand them. As it turns out, the LLMs that many of us use—often through a chat-based interface like ChatGPT—aren't even designed to produce the exact same answer every time. Instead, and I'll confess that I'm simplifying (and perhaps oversimplifying) things here, the models have settings that can introduce variation—“creativity,” as many people call it—into their responses. Again, this is all pretty technical, but the upshot is that an LLM's response reflects its best statistical, probabilistic prediction about the answer to the user's query. See, e.g., Mathew Renze & Erhan Guven, The Effect of Sampling Temperature on Problem Solving in Large Language Models 1–2 (June 14, 2024) (unpublished manuscript), https://arxiv.org/pdf/2402.05201. If a model's creativity dial is turned down, its successive responses to identical questions should be very similar, if not perfectly repetitive.⁸ See, e.g., François Chollet, Deep Learning with Python 374–75 (2d ed. 2021). As it turns out, though, many LLMs’ default creativity settings are dialed up—which, in a nutshell, can subtly adjust the underlying probability distribution that drives the models’ answers. All of which is to say, the very thing that I was concerned about—the LLMs’ subtly different responses, particularly in structure and phrasing—was a feature, not a bug. It's how the models are meant to perform—at least for those of us who (perhaps unwittingly) are using their default settings.⁹

Wait, but why? Why are LLMs’ default parameters set to introduce subtle variation? Doesn't that seem weird, or even a little suspicious? Initially, I certainly thought so—I prefer my data nice and neat, crisp and clean. But upon reflection, I shouldn't have been surprised. Think about how and for what purposes most people—which is to say, not judges—currently use LLMs. If you're brainstorming business ideas, summarizing large masses of information, or (dare I say) “writing” a term paper, then variety is, so to speak, the spice of life. Mindless repetition is bad; a little imagination is good. So, while I'll confess that I find it a little frustrating that I can't explain, at a granular technical level, exactly why the models changed their successive responses, however slightly, it now makes sense to me that they did so. Again, that's how they're designed.¹⁰

B

Now, and even more interestingly to me, the upshot for ordinary-meaning interpretation: Technical explanations aside, why, in retrospect, should I have been so unnerved by the models’ slightly (though again, not substantively) different answers? Given the claims that I've made about LLMs’ utility to the interpretive enterprise, shouldn't a little variation have been exactly what I expected?

Remember, our aim is to discern “ordinary meaning.” Presumably, the ideal gauge of a word's or phrase's ordinary meaning would be a broad-based survey of every living speaker of American English—totally unrealistic, but great if you could pull it off. Imagine how that experiment would go: If you walked out onto the street and asked all umpteen million subjects, “What is the ordinary meaning of ‘physically restrained’?”, I think I can confidently guarantee that you would not get the exact same answer spit back at you verbatim over and over and over. Instead, you'd likely get a variety of responses that differed around the margins but that, when considered en masse, revealed a common core. And that common core, to my way of thinking, is the ordinary meaning.

So, as it turns out, the very thing that had initially given me pause—namely, that the LLMs were returning subtly different responses to the same question—has instead given me (more) hope that the models have something significant to offer the interpretive enterprise. The fact is, language is an organic thing, and like most organic things, it can be a little messy. So too, unsurprisingly, are our efforts to capture its ordinary meaning. Because LLMs are trained on actual individuals’ uses of language in the real world, it makes sense that their outputs would likewise be less than perfectly determinate—in my experience, a little (but just a little) fuzzy around the edges. What's important, though—and I think encouraging—is that amidst the peripheral uncertainty, the LLMs’ responses to my repeated queries reliably revealed what I've called a common core.

Now, I'm not at all sure that we've yet come up with the perfect method for tapping into the models’ seeming ability to quantify and communicate that core. I'm not even sure that existing LLMs (warts and all) can realistically accomplish that goal. But it does strike me that some marginal uncertainty is inherent in the assessment of ordinary meaning, and I (now) view the fact that LLMs appear to reflect that uncertainty as a virtue rather than a vice.

V

So, what are the takeaways?

1. I continue to believe—perhaps more so with each interaction—that LLMs have something to contribute to the ordinary-meaning endeavor. They're not perfect, and challenges remain, but it would be myopic to ignore them.

2. An important (and perhaps underappreciated) benefit of incorporating LLMs into our interpretative practice is their ability to decipher and explain the meaning of composite, multi-word phrases in a way that standard tools like dictionaries can't always do.

3. We should give careful thought to how we assess and account for LLMs’ sometimes varying answers to user queries. For reasons I've explained, I think (a) that there's a valid technical explanation for that phenomenon and (b) that the peripheral variation reflects everyday speech patterns, and thus may actually make the models more (rather than less) accurate predictors of ordinary meaning.

4. A final coda: No one should mistake my missives for a suggestion that AI can bring scientific certainty to the interpretive enterprise. As I've been at pains to emphasize, I'm not advocating that we give up on traditional interpretive tools—dictionaries, semantic canons, etc. But I do think—and increasingly so—that LLMs may well serve a valuable auxiliary role as we aim to triangulate ordinary meaning.

Again, just my two cents. I remain happy to be shouted down.

APPENDIX

ChatGPT

Query: What is the ordinary meaning of “physically restrained”?

Date of query: August 9, 2024

Response 1:

The term “physically restrained” generally refers to the act of limiting or controlling someone's physical movement or freedom through the use of physical force, mechanical devices (such as handcuffs or straps), or by holding them in place. This can occur in various contexts, such as in law enforcement, healthcare, or even everyday situations where someone is being prevented from moving freely. The goal of physical restraint is usually to prevent harm to the restrained person or to others, but it is often considered a serious measure and is typically used as a last resort.

Response 2:

The term “physically restrained” generally refers to the act of using physical force or a device to limit or control someone's movement or actions. This can include being held down, tied up, or confined in a way that restricts the ability to move freely. Physical restraint is often used in contexts like law enforcement, healthcare, or care-giving, where it might be necessary to prevent harm to the person being restrained or to others.

Response 3:

The term “physically restrained” generally refers to the act of limiting or controlling someone's physical movements or freedom by using physical force or devices. This can involve methods like tying someone down, using handcuffs, holding someone back with hands or arms, or employing mechanical devices such as straps or belts to prevent a person from moving freely. It is often used in contexts such as law enforcement, medical care, or safety situations where there is a need to prevent harm to the individual or others.

Response 4:

The ordinary meaning of “physically restrained” refers to the act of limiting or restricting someone's movement through physical means. This could involve the use of physical force, devices like handcuffs or straps, or even holding someone in place with one's hands. The purpose of physical restraint is typically to control or prevent a person from moving freely, often in situations where they may pose a danger to themselves or others.

Response 5:

The term “physically restrained” generally refers to the act of limiting or restricting someone's physical movement through the use of physical force, devices, or restraints. This can involve anything from holding someone down with one's hands, to using handcuffs, straps, or other mechanical restraints to prevent them from moving freely.

The context in which the term is used can influence its precise meaning, but it typically conveys the idea that a person's ability to move or act is being intentionally controlled or limited by another person or mechanism.

Response 6:

The ordinary meaning of “physically restrained” refers to the act of limiting or controlling someone's movement by using physical force or devices. This could involve using handcuffs, straps, or simply holding someone down to prevent them from moving freely. The term is often used in legal, medical, or security contexts to describe situations where a person is being restricted from moving or acting on their own accord.

Response 7:

The ordinary meaning of “physically restrained” refers to the act of limiting or preventing someone's physical movement through the use of physical force, devices, or barriers. This could involve the use of handcuffs, straps, or simply holding someone in place to prevent them from moving freely. The term is often used in legal, medical, and security contexts to describe situations where a person is being controlled or confined to prevent harm to themselves or others.

Response 8:

The ordinary meaning of “physically restrained” refers to a situation where a person's physical movement is restricted or controlled by external forces. This could involve the use of physical barriers, devices (like handcuffs or straps), or the physical presence of another person who holds or confines the individual to limit their ability to move freely. The term is often used in contexts where someone is being held back, prevented from moving, or otherwise kept under control through physical means.

Response 9:

The ordinary meaning of “physically restrained” refers to the act of limiting or controlling a person's movements through physical means. This can involve the use of physical force, devices, or other methods to prevent someone from moving freely. For example, this could include being held down by another person, being tied or bound with ropes or handcuffs, or being confined in a space where movement is restricted. The term generally implies that the person is unable to move or act according to their own will due to these physical limitations.

Response 10:

“Physically restrained” generally means being prevented from moving freely due to physical barriers or constraints. This could involve being tied up, handcuffed, or otherwise restricted in movement by physical means. It's often used in contexts involving security, medical care, or legal situations where someone's movement is limited to control or protect them.

* * *

Claude

Query: What is the ordinary meaning of “physically restrained”?

Date of query: August 9, 2024

Response 1:

The ordinary meaning of “physically restrained” typically refers to the use of physical force or barriers to limit or control a person's movements or actions. This can include:

1. Holding someone back with one's hands or body

2. Using devices like handcuffs, straps, or restraints

3. Confining someone to a specific space, such as a room or cell

4. Preventing someone from moving freely by blocking their path

The term implies a direct, tangible interference with a person's ability to move or act as they wish. It's often used in contexts related to law enforcement, healthcare, or situations where someone's freedom of movement needs to be restricted for safety or security reasons.

The exact interpretation may vary slightly depending on the context, but generally, it involves a clear, physical limitation on someone's freedom of movement. Would you like me to elaborate on any specific aspect of this definition?

Response 2:

The ordinary meaning of “physically restrained” typically refers to:

1. The use of physical force or mechanical devices to limit or control a person's bodily movements.

2. Preventing someone from moving freely by holding them, tying them up, or confining them in a space.

3. Applying physical measures to restrict an individual's ability to act or move as they wish.

This term is often used in contexts such as:

• Law enforcement situations where officers may need to control a suspect

• Medical settings, where patients might need to be restrained for their own safety

• Instances of assault or abduction

• Certain therapeutic or protective interventions

The concept generally implies a temporary and situational application of physical control, rather than a permanent condition.

Would you like me to elaborate on any specific aspect of this definition or its applications?

Response 3:

The ordinary meaning of “physically restrained” typically refers to the act of using physical force or methods to restrict or control a person's movements or actions. This can include:

1. Holding someone back with hands or arms

2. Using restraints like handcuffs, straps, or rope to limit movement

3. Confining someone to a specific space, such as a room or cell

4. Using one's body weight to pin someone down

5. Employing devices designed for restraint, such as strait jackets or leg shackles

The term implies a direct, tangible limitation on a person's physical freedom, usually against their will or as a means of control. It's often used in contexts related to law enforcement, medical settings, or situations where someone's movements need to be restricted for safety or security reasons.

Would you like me to elaborate on any specific aspect of this definition or its applications?

Response 4:

The ordinary meaning of “physically restrained” typically refers to the use of physical force, barriers, or implements to restrict or control a person's bodily movements or actions. This could include:

1. Holding someone back with hands or arms

2. Using restraint devices like handcuffs, straps, or rope

3. Confining someone to a limited space, such as a room or cell

4. Preventing someone from moving freely by blocking their path

5. Using one's body weight to immobilize another person

The term generally implies a deliberate action to limit someone's physical freedom of movement, often against their will or as a means of control. This could occur in various contexts, such as law enforcement, medical settings, or during conflicts between individuals.

It's important to note that the specific interpretation may vary slightly depending on the context in which the term is used. Would you like me to elaborate on any particular aspect of this definition?

Response 5:

The ordinary meaning of “physically restrained” typically refers to the act of using physical force or barriers to limit or control someone's movements or actions. This can include:

1. Holding someone back with your hands or body

2. Using restraints like handcuffs, straps, or rope to restrict movement

3. Confining someone to a specific space, such as a room or cell

4. Using physical objects to block or impede movement

The term implies a direct, tangible intervention to prevent a person from moving freely or taking certain actions. It's often used in contexts like law enforcement, healthcare (particularly in psychiatric settings), or in describing forceful interpersonal interactions.

It's important to note that the use of physical restraints can have legal, ethical, and safety implications depending on the context and manner in which they are applied.

Would you like me to elaborate on any specific aspect of this definition or its applications?

Response 6:

The ordinary meaning of “physically restrained” typically refers to the use of physical force or mechanical means to limit or control a person's movements or actions against their will. This can include:

1. Holding someone back with hands or arms

2. Using handcuffs, straps, or other restraining devices

3. Confining someone to a specific space, like a room or chair

4. Blocking someone's path to prevent movement

The term implies a direct, tangible intervention to restrict a person's physical freedom of movement. It's often used in contexts like law enforcement, healthcare (particularly in psychiatric settings), or in describing forceful interpersonal interactions.

It's important to note that the specific interpretation can vary slightly depending on the context in which it's used. Would you like me to elaborate on any particular aspect or application of this term?

Response 7:

The ordinary meaning of “physically restrained” typically refers to:

1. The use of physical force or mechanical means to limit a person's movement or ability to act freely.

2. Holding someone back or preventing them from moving using bodily contact or restraining devices.

3. Restricting an individual's physical mobility against their will.

This can include actions such as:

• Holding someone's arms or body to prevent them from moving

• Using handcuffs, straps, or other restraining devices

• Confining someone to a specific space, like a room or cell

• Forcibly keeping someone in a particular position

The term is often used in contexts related to law enforcement, healthcare settings, or situations where someone's freedom of movement needs to be limited for safety or security reasons. However, it can also be used in broader contexts to describe any situation where someone's physical movement is forcibly restricted.

Would you like me to elaborate on any specific aspect of this definition?

Response 8:

The ordinary meaning of “physically restrained” typically refers to the use of physical force, barriers, or implements to restrict or control a person's movement or actions against their will. This can include:

1. Using bodily force to hold someone in place

2. Applying handcuffs, straps, or other restraining devices

3. Confining someone to a restricted space, like a locked room

4. Using physical barriers to prevent movement

5. Forcibly limiting someone's range of motion

The term implies a direct, tangible interference with a person's ability to move freely, often in the context of law enforcement, medical situations, or instances where someone's actions need to be controlled for safety reasons.

It's important to note that the specific interpretation may vary slightly depending on the context in which the term is used. Would you like me to elaborate on any particular aspect of this definition?

Response 9:

The ordinary meaning of “physically restrained” typically refers to the act of using physical force or mechanisms to limit or control a person's movement or actions. This can include:

1. Holding someone back with one's hands or body

2. Using devices like handcuffs, straps, or restraints

3. Confining someone to a specific space, such as a room or chair

4. Preventing someone from moving freely by blocking their path

The term implies a direct, tangible intervention to restrict a person's physical mobility or actions, usually against their will or as a means of control. This could occur in various contexts, such as law enforcement, medical settings, or during conflicts.

It's important to note that the specific interpretation may vary slightly depending on the context in which the term is used. Would you like me to elaborate on any particular aspect of this definition?

Response 10:

The ordinary meaning of “physically restrained” typically refers to the act of using physical force or barriers to restrict or limit someone's movement or actions. This can involve:

1. Holding someone back with your hands or body

2. Using restraints like handcuffs, straps, or ropes to immobilize a person

3. Confining someone to a specific space, such as a room or cell

4. Using physical objects to block or impede movement

The term is often used in contexts like law enforcement, medical settings, or situations where someone needs to be prevented from harming themselves or others. It implies a direct, hands-on approach to limiting a person's physical freedom of movement, as opposed to other forms of restraint like verbal commands or legal restrictions.

Would you like me to elaborate on any specific aspect of this definition or provide examples of its use in different contexts?

* * *

Gemini

Query: What is the ordinary meaning of “physically restrained”?

Date of query: August 9, 2024

Response 1:

Physically restrained means that a person's freedom of movement is restricted by another person or object against their will. This can be done in a variety of ways, such as:

• Holding: Someone physically preventing another person from moving.

• Tying: Using ropes, chains, or other materials to bind a person.

• Locking: Confining a person in a space with no means of escape.

Essentially, it means a person is being physically controlled and prevented from moving freely.

Would you like to know the meaning of “physically restrained” in a specific context, such as legal or medical?

Response 2:

Physically restrained means that a person's freedom of movement is restricted by another person or object against their will. This can be done through:

• Direct physical force: Holding, grabbing, or pushing someone to prevent them from moving.

• Physical barriers: Confining someone in a room or space with locked doors or windows.

• Mechanical devices: Using handcuffs, shackles, or similar restraints.

Essentially, it means someone is being held back or prevented from moving freely.

Would you like to know more about the legal or specific context in which this term is used?

Response 3:

Physically restrained means that a person's freedom of movement is restricted by another person or object against their will. This can be done in a variety of ways, such as:

• Holding: Someone physically preventing another person from moving.

• Tying: Using ropes, chains, or other materials to bind a person.

• Locking: Confining a person in a space with limited or no exit.

Essentially, it means someone is being held against their will and unable to move freely.

Would you like to know more about the legal or specific context of this term?

Response 4:

Physically restrained means that a person's freedom of movement is restricted by another person or object against their will. This can be done in a variety of ways, such as:

• Holding: Someone physically holding another person to prevent them from moving.

• Tying: Using ropes, chains, or other materials to bind a person.

• Confining: Locking someone in a room or space.

• Immobilizing: Using physical force or devices to prevent movement.

Essentially, it involves limiting a person's ability to move freely through physical means.

Would you like to know the legal definition of physically restrained, or do you have a specific context in mind?

Response 5:

Physically restrained means that a person's freedom of movement is restricted by another person or object against their will. This can be done through:

• Direct physical force: Holding someone down, pushing them back, or preventing them from moving.

• Physical barriers: Confining someone in a room, tying them up, or using handcuffs.

Essentially, it means someone is being forcibly prevented from moving freely.

Example: A person who is handcuffed to a chair is physically restrained.

Would you like to know the legal definition of “physically restrained” in a specific context?

Response 6:

Physically restrained means that a person's freedom of movement is restricted by another person or object against their will. This can be done through:

• Direct physical force: Holding someone down, pushing them back, or preventing them from moving.

• Physical barriers: Confining someone in a locked room, using handcuffs, or tying someone up.

Essentially, it means someone is being physically prevented from doing what they want to do.

Would you like to know the legal definition in a specific context?

Response 7:

Physically restrained means that a person's freedom of movement is restricted by another person or object against their will. This can involve:

• Holding or grabbing: Using physical force to prevent someone from moving.

• Tying or binding: Using ropes, chains, or similar items to restrict movement.

• Confining: Keeping someone in a limited space, such as a room or cage.

Essentially, it means someone is being physically prevented from doing what they want to do.

Would you like to know about the legal or medical implications of physical restraint?

Response 8:

Physically restrained means that a person's freedom of movement is restricted by another person or object against their will. This can be done through:

• Direct physical force: Holding someone down, pushing them back, or preventing them from moving.

• Physical barriers: Confining someone in a locked room, using handcuffs or shackles, or tying someone up.

Essentially, it involves limiting someone's ability to move freely and independently.

Would you like to know the legal definition or the term's use in a specific context?

Response 9:

Physically restrained means that a person's freedom of movement is restricted by another person or object against their will. This can be done through:

• Direct physical force: Holding, grabbing, or pinning someone down.

• Physical barriers: Using handcuffs, zip ties, or other devices to restrict movement.

• Environmental restrictions: Confining someone in a locked room or space.

Essentially, it means someone is being forcibly prevented from moving freely.

Would you like to know more about the legal or specific context of the term?

Response 10:

Physically restrained means that a person's freedom of movement is restricted by another person or object against their will. This can involve:

• Holding: Someone physically holding another person to prevent movement.

• Tying: Using ropes, chains, or other materials to bind a person.

• Locking: Confining a person in a space with no means of escape.

Essentially, it means a person is prevented from moving freely by external forces.

* * *

FOOTNOTES

1.   As the Majority Opinion notes, we relied in part on the commentary to section 2B3.1(b)(4)(B) when we adopted the broad definition of “physically restrained” in Jones and Victor. Yet under Dupree, because the guideline itself is unambiguous, we shouldn't have gotten to the commentary. We issued United States v. Ware, 69 F.4th 830 (11th Cir. 2023), after Dupree, so perhaps we could have corrected this aspect of our precedent in Ware. But we didn't do that. Of course, “when circuit authority is in conflict, a panel should look to the line of authority containing the earliest case, because a decision of a prior panel cannot be overturned by a later panel.” MacPhee v. MiMedx Grp., Inc., 73 F.4th 1220, 1250 (11th Cir. 2023) (citation and quotation marks omitted). But we can't follow Dupree instead of Ware now as the earlier-issued case because Ware’s holding—though arrived at in a way contrary to Dupree’s stated methodology—does not conflict with Dupree’s holding. See Washington v. Howard, 25 F.4th 891, 900 (11th Cir. 2022) (explaining that we must reconcile prior panel decisions whenever possible, and that “only the holdings of prior decisions bind us” and “legal principles set forth outside of the decision's holding do not bind us”).

2.   Restrain, Oxford English Dictionary, https://www.oed.com/dictionary/restrain_v1?tab=meaning_and_use#25713964 [https://perma.cc/9FW3-5UJ2].

3.   Restrain, Merriam-Webster Dictionary, https://www.merriam-webster.com/dictionary/restrain [https://perma.cc/TV96-CHJ8].

4.   Physically, Oxford English Dictionary, https://www.oed.com/dictionary/physically_adv?tab=meaning_and_use#30459183 [https://perma.cc/S4BN-FN6D].

5.   Physically, Merriam-Webster Dictionary, https://www.merriam-webster.com/dictionary/physically [https://perma.cc/6263-7SBF].

6.   Id.

7.   Physical, Oxford English Dictionary, https://www.oed.com/dictionary/physical_adj?tab=meaning_and_use#30455352 [https://perma.cc/Q7QV-XDDK].

8.   Dupree, 57 F.4th 1269, sets forth the process for interpreting the Guidelines. It relied on the deferential scheme afforded to agency interpretations of their own legislative rules when those rules are truly ambiguous. Id. at 1275–77. Recently, the Supreme Court held that, where Congress has not conferred discretionary authority on an agency to interpret a federal statute, the Administrative Procedure Act, 5 U.S.C. § 706, does not permit courts to “defer to an agency interpretation of the law simply because a statute is ambiguous.” Loper Bright Enters. v. Raimondo, ––– U.S. ––––, 144 S. Ct. 2244, 2263, 2268, 2273, ––– L.Ed.2d –––– (2024). To be sure, that is not squarely on point with Dupree, which addressed the proper interpretation of the Sentencing Guidelines and their commentary and application notes, or the cases Dupree relied on. But to the extent Loper Bright casts any doubt on Dupree’s methodology, that may be another good reason for us to consider rehearing this case en banc.

9.   As the Majority Opinion notes, in United States v. Jones, 32 F.3d 1512 (11th Cir. 1994), we based our decision to so broadly define “physically restrained” on the commentary's use of “such as.” Id. at 1518. We said that phrase indicates that the “illustrations of physical restraint are listed by way of example rather than limitation.” Id. at 1518. I agree that the phrase “such as” means that a person can engage in conduct that warrants the enhancement other than through the modes the commentary lists. But most respectfully, I don't see how the phrase “such as” somehow permits broadening the scope of the guideline beyond “physical” restraint. The doctrine of noscitur a sociis—that is, a word is known by the company it keeps—tells us that a word is “given more precise content by the neighboring words with which it is associated.” Fischer v. United States, ––– U.S. ––––, 144 S. Ct. 2176, 2183, ––– L.Ed.2d –––– (2024) (citation and quotation marks omitted). So while the phrase “such as” denotes that the enhancement contemplates the inclusion of conduct other than that listed, other qualifying conduct must be similar to the conduct that the commentary enumerates. And all that conduct is physical in nature.

10.   Of course, that would not prevent a sentencing court from considering Deleon's gun-pointing conduct in crafting an appropriate sentence; the sentencing court simply could not apply section 2B3.1(b)(4)(B)’s two-point enhancement.

11.   The Third Circuit takes yet another approach when considering whether conduct qualifies as physical restraint. See United States v. Bell, 947 F.3d 49, 56 (3d Cir. 2020) (adopting five factors to balance when deciding whether to impose the enhancement, namely, “[1] Use of physical force; [2] Exerting control over the victim; [3] Providing the victim with no alternative but compliance; [4] Focusing on the victim for some period of time; and [5] Placement in a confined space”).

1.   Worth filing away for another day: The breaking-and-repiecing approach won't always work so well. To be sure, sometimes a phrase really is just the sum of its parts—say, “strictly enforced” or “accurately measured.” But that's not uniformly true. Consider, for instance, common idioms like “kick the bucket” or “spill the beans.” Or metaphors like “falling in love” or “drowning in work.” Or, even beyond those, plain old linguistic combinations like “white noise” or “dark matter.” Most have a sense for what those phrases denote, but their meanings wouldn't be self-evident from splicing together dictionary definitions of each constituent word. “White noise,” for example, isn't a “[l]oud, confused, or senseless shouting or outcry,” or even a “sound of any sort,” that is “[o]f the color of pure snow, milk, or sunlight.” Webster's New International Dictionary 1658, 2915 (2d ed. 1944) (defining both terms). Again, I think Judge Rosenbaum has shown that the breaking-and-repiecing method works well enough for “physically restrained.” My point is simply that it won't always do the trick.In those instances where a phrase is more than the sum of its component parts, I think that LLMs may well help to fill the gaps left by word-centric dictionaries. The reason: They aren't dependent on human beings going to the trouble of identifying and then defining commonly used phrases. There are no doubt others who can explain the technical details better than me, but as I understand things, the models’ ability to tackle this task—i.e., to make predictions about the meaning of phrases, in context—is hard-wired into their DNA, inherent in what they're designed to do. See Snell, 102 F.4th at 1226–30 (Newsom, J., concurring); see also Yonathan Arbel & David A. Hoffman, Generative Interpretation, 99 N.Y.U. L. Rev. 451, 479 (2024) (explaining how the attention mechanism helps LLMs understand phrases like “he swung the bat”). Why? Well, consider several of the uses to which people frequently put these models in professional settings: (1) summarizing information contained in books or articles; (2) brainstorming ideas for a business meeting; and (3) providing feedback on written work. Cf. Kevin Roose, How Should I Use AI Chatbots Like ChatGPT?, N.Y. Times (Mar. 30, 2023), https://www.nytimes.com/2023/03/30/technology/ai-chatbot-chatgpt-uses-work-life.html. In order for an LLM's output to be useful to everyday consumers, the model has to be able to understand and distinguish between the different uses of a phrase like, say, “passed out.” Suffice it to say, an LLM that can't ascertain that “passed out” means something different in the sentences (1) “The teacher passed out the writing assignment” and (2) “The teacher passed out on the floor” wouldn't be very useful. And yet—and this is the important point—the models can (with uncanny accuracy) draw that distinction because they've crunched enough raw linguistic data to “understand,” based on contextual clues, how the phrase is being used in a particular circumstance.Anyway, perhaps I'll revisit this issue down the road, in a case in which the breaking-and-repiecing method of explaining a compound phrase's meaning comes up short.

2.   To be more precise, I queried the free online version, which uses GPT4-o.

3.   Specifically, Claude 3.5 Sonnet—again, the free version.

4.   Gemini 1.5 Flash—which, again, is free.

5.   Because I used the free chat-based interfaces available online, I couldn't (so far as I know) adjust the LLMs’ “temperature” settings as some have done in more formal studies. See, e.g., Arbel & Hoffman, supra, at 487–488; Engel & McAdams, supra, at 15–16. (“Temperature” is a setting that can influence the predictability—or alternatively, “creativity”—of a model's responses.) I opted for the less technical approach because I think it's worth exploring how LLMs can be used in a low-cost, user-friendly way—and thereby, I hope, best approximate how judges, lawyers, and ordinary citizens might typically employ them.Having said that, I was able—check that, one of my genius law clerks was able—to figure out how to use the (more sophisticated, and paywalled) Open AI Playground. He asked the same question, “What is the ordinary meaning of ‘physically restrained’?”, and he set the temperature to zero. By doing so, we were able, I think, to get even a wee bit closer to the model's “best guess” at the answer to my question. See Engel & McAdams, supra, at 40 n.128. Playground's response basically tracked ChatGPT's:The ordinary meaning of “physically restrained” refers to the act of limiting or controlling someone's movements through physical means. This can involve using physical force, devices, or other methods to prevent a person from moving freely. Examples include handcuffs, ropes, straps, or simply holding someone down with one's hands. The key aspect is that the restriction is achieved through physical measures rather than verbal commands or psychological tactics.

6.   I realize that my mini-experiment differed from others’ empirical work in another respect. Rather than asking about a word's or phrase's “ordinary meaning,” as I did, those studies (at least in part) asked LLMs to apply their predictions about ordinary meaning to a case's facts and then render final decisions—e.g., In the circumstances presented by Deleon's case, “Was anyone physically restrained?” See, e.g., Arbel & Hoffman, supra, at 487; Engel & McAdams, supra, at 25. As I've said before, nailing down the “proper question” is critically important. See Snell, 102 F.4th at 1232–33 (Newsom, J., concurring). For better or worse, I tend to default to asking, more generally, for a term's ordinary meaning—probably because that's what I'm doing when I consult dictionaries. If there's a better way—a better question—I'm fine with that.

7.   So, for instance, Gemini often mentioned “holding, tying, and locking” as examples of physical restraints.

8.   I'm thinking here principally about an LLM's “temperature” setting, which appears to be the most frequently discussed creativity parameter. See supra at –––– – –––– n.5.

9.   I'll confess that I couldn't find an authoritative explanation of what the leading LLMs’ default creativity settings are for those of us who interact with the models via a chat-based interface like ChatGPT. Based on the subtle variation I saw, though, I'm assuming they're not zero—i.e., that the models are introducing some variation into their responses. And it seems reasonable to assume, as well, that the settings are similar for those who plug into the models directly. See, e.g., Anthropic, Create a Text Completion, API Reference, https://docs.anthropic.com/en/api/complete [https://perma.cc/KHX3-5PKA] (last visited Sept. 4, 2024) (explaining that temperature “[d]efaults to 1”).

10.   One more thing I should mention briefly before moving on: Without thinking about it one way or the other, when I initially queried Claude for the second time, I did so within the same chat “string.” I'm afraid, therefore, that I might have unwittingly tapped into the chatbot's “conversational history.” See, e.g., Akash Gupta et al., LLM Task Interference: An Initial Study on the Impact of Task-Switch in Conversational History 1–2 (Feb. 28, 2024) (unpublished manuscript), https://arxiv.org/pdf/2402.18216. The basic idea, so far as I understand it—I keep feeling the need to say that—is that an LLM will adjust its behavior based on the questions the user feeds it, offering clarifications and refinements along the way. (Rather than claiming that the model is “learning” from repeated questions, I think it's probably safer and more accurate to say that its answers are becoming more “precise” or “useful.” An LLM's knowledge base is essentially fixed, based on the underlying data on which it trains; it doesn't “update” in response to questions.) Indeed, figuring out how best to use prompts to better guide LLMs to more accurate results is a new and valuable skill. See, e.g., Jack Kelly, The Hot, New High-Paying Career Is an AI Prompt Engineer, Forbes (Mar. 6, 2024), https://www.forbes.com/sites/jackkelly/2024/03/06/the-hot-new-high-paying-career-is-an-ai-prompt-engineer/ [https://perma.cc/45VA-YJC4]. In any event, as explained in text, I tweaked my approach for my mini-experiment, opening a new chat for each successive question.

Rosenbaum, Circuit Judge: